LIVS: A Pluralistic Alignment Dataset for Inclusive Public Spaces

Rashid Mushkani, Shravan Nayak, Hugo Berard, Allison Cohen, Shin Koseki, Hadrien Bertrand

TL;DR: This paper presents LIVS, a dataset of diverse community feedback on public space design, used to fine-tune Stable Diffusion XL. Results show moderate alignment shifts while highlighting persistent subjectivities in AI-generated urban imagery.

Abstract: We introduce the *Local Intersectional Visual Spaces* (LIVS) dataset, a benchmark for multi-criteria alignment, developed through a two-year participatory process with 30 community organizations to support the pluralistic alignment of text-to-image (T2I) models in inclusive urban planning. The dataset encodes 37,710 pairwise comparisons across 13,462 images, structured along six criteria—Accessibility, Safety, Comfort, Invitingness, Inclusivity, and Diversity—derived from 634 community-defined concepts. Using Direct Preference Optimization (DPO), we fine-tune Stable Diffusion XL to reflect multi-criteria spatial preferences and evaluate the LIVS dataset and the fine-tuned model through four case studies: (1) DPO increases alignment with annotated preferences, particularly when annotation volume is high; (2) preference patterns vary across participant identities, underscoring the need for intersectional data; (3) human-authored prompts generate more distinctive visual outputs than LLM-generated ones, influencing annotation decisiveness; and (4) intersectional groups assign systematically different ratings across criteria, revealing the limitations of single-objective alignment. While DPO improves alignment under specific conditions, the prevalence of neutral ratings indicates that community values are heterogeneous and often ambiguous. LIVS provides a benchmark for developing T2I models that incorporate local, stakeholder-driven preferences, offering a foundation for context-aware alignment in spatial design.

Lay Summary:

Public spaces such as parks and streets shape daily life in cities, but the way these places are designed is often driven by professional or financial interests. This means that people who use these spaces, especially those from marginalized groups, may not have much influence over how they look or function. Community members who want to participate in design processes face barriers like inaccessible venues, scheduling issues, and visual tools that do not reflect social or sensory needs.

To address this, we worked with 30 community organizations in Montréal over two years to collect local perspectives on public space. Together, we identified six important criteria for inclusive design: accessibility, safety, comfort, invitingness, inclusivity, and diversity. Community members compared images of parks, streets, and plazas, producing over 37,000 annotations that reflect a wide range of experiences.

Using these annotations, we adapted an open-source text-to-image AI model. We trained the model to better represent local preferences by fine-tuning it with the community-generated feedback. When tested, the new model sometimes reflected community priorities more closely, especially for criteria that had more training data. However, many participants found it difficult to choose between options, especially for concepts like inclusivity or diversity. This suggests that people’s needs are varied and not always easy to represent in one system. Our approach provides a way for residents and planners to explore design choices together, but it does not remove the challenges that come from competing priorities within a community.

Application-Driven Machine Learning: This submission is on Application-Driven Machine Learning.

Link To Code: https://mid-space.one/

Primary Area: Applications->Social Sciences

Keywords: Pluralistic Alignment, Text-to-Image Diffusion, Intersectionality, Urban Planning, DPO, Inclusivity, Safety, Accessibility

Submission Number: 11719